CONCLUSION
In this thesis, we first attempted to validate the findings by Singh et al. (2015) on a transactional database provided by a different major bank in the same OECD country. We implemented the same features defined and introduced by Singh et al. (2015) and applied the same bagging model for predicting three output indicators of financial well-being: overspending, trouble due to bank’s administrative action, and late payment.
The results of our research validate the significance of spatio-temporal mobility features over demographic features in predicting an individual’s propensity to ‘overspend’ or be in financial ‘trouble’. In the case of predicting whether a customer is bound to miss their payments or not, the demographic features of the individual helped us predict to a better accuracy. Specifically, spatio-temporal features predicted ‘overspending’ and ‘trouble’ better than demographic features by around 6% and 9% respectively. Whereas the demographic features predicted ‘late payment’ by at least 5% better than the spatio- temporal features.
In the second part of the thesis, we introduced new input variables designed as behavioral features using the same dataset provided by A-Bank. Here we considered shopping categories and banking channels as new measures depicting customer behavior. The results with this new set of features suggest that considering the most frequented shopping category helped in predicting ‘overspending’ better than all the other features. Precisely, it predicted ‘overspending’ better than the spatio-temporal features by at least 14%. This proves that an individual’s behavior, which includes both his/her spatio-temporal mobility and purchasing behavior, can significantly predict whether they will overspend or not. These findings can be considered to be of high importance since they have the ability to affect the markets of credit repayment and credit card limit decisions worth over a trillion dollars, (Singh, Bozkaya, & Pentland, 2015).
In addition to the entropy measures and binary indicator variables, there are other types of variables or features which could be considered as inputs to the prediction model. For example, ‘age’ can be treated as an interval variable and several age interval groups (e.g. age 19-25, 26-35) can be created out of the data. In addition, data regarding customer’s banking age, product ownership of equity- and debt-related products, and even the number of credit cards owned could be experimented with to see if any such association can be found with the financial outcomes.
An important point to remember is that this research was carried out on a dataset belonging to a single major metropolitan city of the OECD country. Big-city residents have the propensity to travel long distances both within the city and outside of it. Hence, for other cities the mobility and financial behavior dynamics can be quite different. For example, residents of small cities may not have great variation in mobility overall and may prefer to travel very less distances within the city due to the proximity of places such as stores and bank branches. Another issue is the possible lack of point-of-sale (POS) and
other mobility-tracking devices, which might not be available say at the local butcher or clothing shop. Hence, this can affect the range of data available to generate behavioral features out of them. Furthermore, due to a lower level of education and income, small- city residents may not even be aware or prefer to own a credit card due to their aversion to risk (Tavor & Garyn-Tal, 2016), which eventually affects the level of credit card transaction information available for data modeling.
In addition, there are several factors which can affect the validity of the model presented in this research. Firstly, dataset differences in terms of customer profiles can affect the findings. For example, a younger-aged dataset may have a higher percentage of over- spenders and late payers, as compared to an older-aged dataset. Also, a dataset where the average customer income is quite low may not show a lot of variation in mobility and may also be very regular and loyal in behavior. In terms of the types of banks, where a private bank encourages lending and charges higher interest rates, a public bank will discourage lending but charge low interest rates. This can have implications for the spending behavior of customers: the dataset of a private bank may have high diversity in both shopping behavior and even the shopping categories which are preferred, due to the easy availability of credit. This may even lead to a high percentage of people who are in financial trouble, over-spending, and paying their dues late. On the other hand, the dataset of a public bank may have high regularity in terms of shopping behavior and have low number of people who are in financial trouble, overspending, or paying their dues late. And finally, as transactions are moving online, the validity of mobility-based features to predict financial outcomes can be threatened, and hence there can be a shift to using other types of variables. These might then be related more to temporal features, or even to demographics such as age and marital status.
Overall, the availability of an individual’s transactional profile now allows us to create personalized models which can be used to predict their behavior. The modeling techniques and analysis described in this thesis can even be done in advance of the individual’s credit card payment deadlines, hence allowing banks to take preemptive measures before the customer gets into ‘trouble’ or ‘overspends’. As financial data arising from POS machines and ATMs increasingly become geo-coded, it is expected that the information related to a person’s spatio-temporal mobility and spending habits will become increasingly available, allowing us to create more robust and accurate models.
Individuals may even use the observations and information presented in this research to rectify their own behavior before they display any of the negative financial outcomes. For example, if a user is becoming increasingly regular in the shopping locations which they visit or in the day of the week they shop at, he/she can be notified that they have a high risk of overspending. This will allow the user to modify and become conscious of their behavior. However, the user may or may not wish to take heed of this information and may either ignore the message altogether or use it to rectify their behavior.
In the end, a side piece of information albeit an important one is that all these findings are based on an individual’s mobility which can have implications on their privacy. It is imperative that users become aware of the level of information which they share, both consciously and unconsciously. In a Bit9 report (2012), it was discovered that more than 40% of mobile applications ask permission to use the location of the user (Sverdlove & Cilley, 2012), and according to Wired.com, some of these applications such as a flashlight do not even need such level of information to function, (McMillan, 2014). Hence, to not become a victim to a breach of privacy, one should be conscious of which applications and systems one has allowed for tracking.
REFERENCES
Achtziger, A., Hubert, M., Kenning, P., Raab, G., & Reisch, L. (2015). Debt out of control: The links between self-control, compulsive buying, and real debts. Journal of Economic Psychology, 141-149.
Christiansen, J., Fatnani, S., Kolhatkar, J. S., & Srinivasan, K. (2001). Washington, DC Patent No. US 6,202,053.
Constangiora, A. (2011). Consumer Credit Scoring. Romanian Journal of Economic Forecasting, 3, 162-177.
Corsetti, G., Pesenti, P., & Roubini, N. (1999). What caused the Asian currency and financial crisis? Japan and the world economy, 305-373.
Crook, J. N., Edelman, D. B., & Thomas, L. C. (2007). Recent developments in consumer credit risk assessment. European Journal of Operational Research, 183(3), 1447- 1465.
Dhar, R., & Wertenbroch, K. (2000). Consumer Choice Between Hedonic and Utilitarian Goods. Journal of Marketing Research, 60-71.
Dong, X., Suhara, Y., Bozkaya, B., Singh, V. K., Lepri, B., & Pentland, A. S. (2018). Social Bridges in Urban Purchase Behavior. ACM Transactions on Intelligent Systems and Technology (TIST), 9(3), 33.
Gonzalez, M. C., Hidalgo, C. A., & Barabasi, A. L. (2008). Understanding individual human mobility patterns. Nature, 453(7196), 779.
Hand, D. J. (2001). Modelling consumer credit risk. IMA Journal of Management Mathematics, 12(2), 139-155.
Hand, D. J., & Henley, W. E. (1997). Statistical Classification Methods in Consumer Credit Scoring: a review. Journal of the Royal Statistical Society: Series A (Statistics in Society), 160(3), 523-541.
Henley, W. E., & Hand, D. J. (1996). A k-nearest-neighbour classifier for assessing consumer credit risk. The statistician, 77-95.
Hui-Yi, L., & Nigel, H. (2012, September 25). Effects of shopping addiction on consumer decision-making: Web-based studies in real time. Journal of Behavioral Addictions, 1(4), 162-170.
Isaacman, S., Becker, R., Caceres, R., Kobourov, S., Rowland, J., & Varshavsky, A. (2010). A tale of two cities. HotMobile '10 Proceedings of the Eleventh Workshop on Mobile Computing Systems & Applications (pp. 19-24). Annapolis, Marland: ACM New York.
Khandani, A. E., Kim, A. J., & Lo, A. W. (2010). Consumer credit-risk models via machine-learning algorithms. Journal of Banking & Finance, 34(11), 2767-2787. Khashman, A. (2010). Neural networks for credit risk evaluation: Investigation of different neural models and learning schemes. Expert Systems with Applications, 37(9), 6233-6239.
Kruppa, J., Schwarz, A., Arminger, G., & Ziegler, A. (2013). Consumer credit risk: Individial probability estimates using machine learning. Expert Systems with Applications, 40(13), 5125-5131.
Kung, K. S., Greco, K., Sobolevsky, S., & Ratti, C. (2014). Exploring universal patterns in human home-work commuting from mobile phone data. Plos One, 9.
Lazer, D., Pentland, A. S., Adamic, L., Aral, S., Barabasi, A. L., Brewer, D., . . . Jebara, T. (2009). Life in the network: the coming age of computation social science. Science, 323(5915), 721.
McMillan, R. (2014, October 20). THE HIDDEN PRIVACY THREAT OF ...
FLASHLIGHT APPS? Retrieved from Wired:
https://www.wired.com/2014/10/iphone-apps/
Noulas, A., Scellato, S., Lathia, N., & Mascolo, C. (2012). A random walk around the city: New venue recommendation in location-based social networks. SOCIALCOM PASSAT '12 Proceedings of the 2012 ASE/IEEE International Conference on Social Computing and 2012 ASE/IEEE International Conferecne on Privacy, Security, Risk, and Trust (pp. 144-153). Amsterdam: IEEE.
Singh, V. K., Bozkaya, B., & Pentland, A. (2015). Money Walks: Implicit Mobility Behavior and Financial Well-Being. Plos One.
Sobolevsky, S., Sitko, I., Des Combes, R. T., Hawelka, B., Arias, J. M., & Ratti, C. (2014). Money on the move: Big data of bank card transactions as the new proxy for human mobility patterns and regional delineation. the case of residents and foreign visitors in Spain. Big Data (BigData Congress), 2014 IEEE International Congress (pp. 136-143). IEEE.
Sverdlove, H., & Cilley, J. (2012, October). Pausing Google Play: More Than 100,000 Android Apps May Pose Security Risks With Mobile Security Survey. Retrieved
July 7, 2018, from Carbon Black:
https://www.bit9.com/download/reports/Pausing-Google-Play-October2012.pdf Tavor, T., & Garyn-Tal, S. (2016). Further examination of the demographic and social
factors affecting risk aversion. Financial Markets and Portfolio Management, 30(1), 95-110.